Development of numerical optimization algorithms for power system real-time estimation tools

Abstract

The estimation of voltage/current phasors and local frequency are key parts to operate and control the ever evolving electricity networks. Phasor and frequency can be estimated by using numerical optimisation techniques. These techniques offer the flexibility to handle frequency variations, disturbances, and fault conditions more effectively when compared with the commonly used digital techniques such as fast Fourier transform (FFT) and discrete Fourier transform (DFT). Although numerical techniques such as Kalman filter (KF), Newton-type algorithm (NTA), recursive-least-squares (RLS), and least-squares (LS) rectify the shortcomings of FFT and DFT techniques, their immense computational burden and stability issues impede their real-time application. Adaptive notch filter (ANF) and enhanced phase-locked loop (EPLL) benefit from simple structures although their performance degrade under severe unbalance and fault conditions. The objective of this thesis is to propose the application of new numerical optimisation techniques as real-time estimation tools for power system phasors, harmonics, and frequency. The contributions of this research work are as follows:  The LS technique is modified to deliver frequency adaptive estimate of voltage envelope.  A new decoupled recursive-least-squares (DRLS) technique is presented for harmonics and phasor estimation.  A modified gradient search (MGS) undersampling technique for harmonics/interharmonics and phasor estimation is presented which shows comparable accuracy with that of the recommended technique by IEC Standard 61000-4-7.  A new Newton-type-algorithm and least-squares (NTA-LS) frequency estimation technique is presented. MATLAB software has been used to emulate harmonics, frequency variation, and fault conditions in a computer environment. Programmable power supplies have been used to conduct experiments in the laboratory. Two widely available research & development digital signal processor (DSP) controllers have been used for real-time phasor and frequency estimation in the experiments. Simulation studies and real-time experiments have been performed and results presented in this thesis support the following observations.  The proposed LS technique shows better performance in tracking voltage envelope and frequency variations when compared with the ANF and EPLL techniques. The proposed new least-squares-Kalman (LSK) estimation technique shows better accuracy than the DFT technique in tracking voltage flicker patterns.  The DRLS technique shows considerable improvement in terms of reducing the computational burden of the conventional RLS technique for real-time implementation on DSPs. The computational efficiency of the DRLS technique is also better than that of the DFT and adaptive linear combiner (ADALINE) techniques for extracting phasors and harmonics under highly harmonic distorted power system environments. The DRLS technique also shows a higher level of accuracy and stability to track synchronized phasors under fault conditions when compared with the EPLL technique.  The simplicity of the proposed undersampling MGS technique is promising for analysing the voltage and current spectra in real-time on the simple hardware platforms at a low sampling frequency. This thesis also proposes an extension of the staggered undersampling to reduce the estimation delays associated with the staggered undersampling. The presented extension on the undersampling concept lays the basis of wide-area-measurement systems (WAMs) implementation at the distribution level based on what the residential area broadband internet services can already offer. The proposed technique of direct transfer of data over the communication network is more reliable than the global positioning system (GPS) based synchrophasor detection and bypasses the DSP clock synchronization errors.  The NTA-LS technique demonstrates faster estimation pace, higher accuracy and better noise immunity for implementation at low sampling frequency in comparison with the PLL and DFT-based decomposition of single-phase into orthogonal components (DSPOC) techniques. The stability of the proposed modified NTA-LS technique is also superior to that of the conventional NTA. This work lays the basis for using new numerical optimisation techniques for real-time parameter estimation in power systems and studying the electric load based on slow sampling observations.